PyTorch implementation of RRID

This is the implementation of the paper "Relation Network for Person Re-identification (AAAI 2020)".

For more information, checkout the project site [website] and the paper [PDF].

Dependencies

• Python 3.6
• PyTorch >= 0.4.1
• numpy
• h5py

Datasets

Download the datasets into datasets folder, like ./datasets/market1501/

The market1501 dataset is only available now, and other datasets will be uploaded later

• Market1501 [market1501]
• DukeMTMC-ReID [[dukemtmc]()]
• CUHK03 labeled [[cuhk03_labeled]()]
• CUHK03 detected [[cuhk03_detected]()]

Training

git clone https://github.com/cvlab-yonsei/projects
cd projects/RRID/code
python Train.py --gpus 0 1 # for training

• You can freely define parameters with your own settings like

  python Train.py --gpus 0 1 --dataset_path 'your_dataset_directory' --dataset_type market1501 --exp_dir 'your_log_directory'

  Pre-trained model

• Download pre-trained weights
  Link: [weights]
• Two GPUs are needed to implement the trained model
• The version of pytorch must be 0.4.1 when you implement the model with these weights

Evaluation

• Test the model with our pre-trained weights

  python Evaluate.py --gpus 0 1 --pretrained_weights_dir pretrained_weights.pth 

• Test your own model

  python Evaluate.py --exp dir log

Bibtex

@inproceedings{park2019relation,
  title={Relation Network for Person Re-identification},
  author={Park, Hyunjong and Ham, Bumsub},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2020}
}